基于多模态深度学习的糖尿病肾病诊断方法研究

糖尿病肾病（diabetic nephropathy, DN）是糖尿病患者最常见的微血管并发症之一，也是终末期肾脏病的主要病因，在全球成人中发病率高达9%。在临床实际工作中，糖尿病合并肾脏损害患者的临床表现极为复杂，尤其涉及区分鉴别糖尿病肾病与非糖尿病性肾脏疾病，准确诊断面临巨大挑战。从目前国内外研究情况和现状来看，尚未出现理想的关于DN的无创诊断模型。本项目是申请人团队在基金项目前期研究基础上，结合近年来的理论和技术研究、临床实践数据和经验积累，创新性地将临床常规检验检查结果与环境、遗传等多模态因素相结合，通过加入表观遗传学如甲基化等指标和医学影像数据，采用多模态深度学习等新型人工智能算法，改进并完善DN无创鉴别诊断方法与技术，进而开发一个自动诊断原型系统。基于此原型诊断系统，进行大样本、多中心临床验证并进行优化，为DN无创鉴别诊断方法在临床上大规模的应用提供科学依据和技术基础。

Diabetic nephropathy (DN) is one of the most common microvascular complications in diabetic patients and a major cause of end-stage renal disease, with a prevalence of 9% in adults worldwide. In clinical practice, the clinical manifestations of patients with diabetes and kidney damage are extremely complicated, especially in the differential diagnosis of diabetic nephropathy and non-diabetic kidney disease. The accurate diagnosis of them faces enormous challenges. From the current domestic and international research situation, there is no ideal non-invasive diagnostic model for DN. This project is based on the preliminary research of the fund project, combined with recent theoretical and technical research, clinical practice data and experience accumulation, innovatively combining clinical routine test results with environmental, genetic and other multimodal factors. By adding epigenetics such as methylation and other medical imaging data, and using new artificial intelligence algorithms such as multimodal deep learning, the DN non-invasive differential diagnosis method and technology were improved, and an automatic diagnostic prototype system was developed. Based on the prototype diagnostic system, large-sample, multi-center clinical trials will be validated and optimized to further provide a scientific and technical basis for the large-scale clinical application of DN non-invasive differential diagnosis methods.